For translators

Frontier AI labs have a language problem. The corpora their models were trained on are heavily weighted toward English. As models are deployed globally, the labs need to raise their performance in every other language, especially the ones where the training data is thinnest or the professional standard is highest. Filling that gap requires professional translators who know the difference between a technically correct translation and one that sounds like a working professional would actually write.

The demand is disproportionate to the visible supply. If you work as a translator, especially in a language pair involving Arabic, Bengali, Hindi, Urdu, Swahili, Vietnamese, Indonesian, Tagalog, Turkish, Persian, Ukrainian, or one of the other under-served language pairs, the current market for this work is real.

What the work looks like

• Translation evaluation. You are given a source text and a model translation. You judge it against professional standards, flag mistranslations, and mark register or idiomatic errors.
• Adversarial testing. You construct prompts where the model is likely to produce a plausible-sounding but subtly wrong translation, especially at register, idiom, or domain-specific vocabulary.
• Reference translation. You produce human reference translations of specific texts for evaluation datasets.
• Rubric review. You audit the criteria the lab is using to score translations in your language pair and flag anything that would let a common failure mode through.

The work is asynchronous. Blocks of two to four hours. Structured deliverables.

What labs actually look for

Professional translation experience in a specific language pair. Written fluency at the level a professional client would expect. Domain specialization (medical, legal, technical, literary) helps but is not required for general work. Certification matters less than a demonstrated track record.

Where the pay lands

Rates are in USD and are set against Western translation-industry scales, priced by language pair and difficulty. For translators in emerging markets, this typically compares very favorably against local agency rates. Under-served language pairs command a premium.

How to enter

Sign up, upload a recent CV or professional summary showing your working language pairs and any domain specialization, and let the platform surface open translation-evaluation roles. Apply where the fit is honest.